Логин: Пароль:
Регистрация нового пользователя
Модератор: Walter Montego 
 Chess variants (10x8)

Sam has closed his piano and gone to bed ... now we can talk about the real stuff of life ... love, liberty and games such as
Janus, Capablanca Random, Embassy Chess & the odd mention of other 10x8 variants is welcome too


For posting:
- invitations to games (you can also use the New Game menu or for particular games: Janus; Capablanca Random; or Embassy)
- information about upcoming tournaments
- disussion of games (please limit this to completed games or discussion on how a game has arrived at a certain position
... speculation on who has an advantage or the benefits of potential moves is not permitted while that particular game is in progress)
- links to interesting related sites (non-promotional)


Список форумов
Режим: Каждый может объявить
Поиск в сообщениях:  

8. Ноября 2005, 22:53:01
Chicago Bulls 
Walter Montego: A computer can run 24 hours a day studying each of the 960 positions and just keep getting more and more prepared for the next upcoming tournament.

There is something behind that that is very tricky!
Even if we suppose that we let a computer run and play against another computer FRC games, for even 15 years or 50 and then build an opening repertoire from that games, WE SHOULD NOT IN ANY WAY, CONCLUDE THAT THIS OPENING WE HAVE BUILT, IS A DECENT ONE TO PLAY FRC CHESS!

I had to use bold-capitals in that statement because it's a well known fact. The reason that our book that is based on 50 years or 1.000.000.000 games, is not suitable for a strong FRC opening book? Because it is based on the knowledge of the 2 computers they play and any weaknesses these 2 have will be included in the book! Even if we had 10 different computers with 10 different styles (personalities) for 50 years to play, then again we do nothing at all! Zero! Even if we include book learning (there is such an option now for the record) in the computers.
Again because the positional (mainly) weaknesses would be a major factor for being our opening book bad.
And these weaknesses will be exposed if we allow after 50 years the computer to play against a human GM at FRC........
So to build a good FRC book we need to play by both humans AND computers for a very long time AND successive learning by both of them during the process. This last one is very important........

For example such questions occur very often to Backgammon, where the top playing programs have obtained their enormous strength by playing millions of games against itself. Strangely enough this approach works for Backgammon while at Chess fails miserably......

The procedure is simple:
Program a Backgammon engine with some simple rules of knowledge. This engine would be a complete moron at Backgammon. Let it play 10.000 games against itself. Let it learn from its mistakes and then import the knowledge into a new engine. Let this engine play another 10.000 games and repeat the procedure. After many learning stages and about 60.000 games you will have a very strong Backgammon bot. This doesn't work at Chess!

And i say it is odd to reach their enormous strength by playing games against itself because this procedure would seeminly lead as i said to playing engine that would have many weaknesses in some areas. But it doesn't! Actually it does in some technical plays of Backgammon bots, but it's not so significant to prevent them plat at top level. But at Chess it fails completely!

Дата и время
Друзья в сети
Любимые форумы
Клубы
Советы
Копирайт © 2002 - 2024 Филип Рахунек, все права зарезервированы.
Наверх